Signal characteristic extractio of wood defects based on wavelet packet

The wavelet packet decomposition was made for the ultrasonic testing signal of wood defects. The wavelet function of Db5 was applied to make the three-layer wavelet packet decomposition for the wood defects. Four characteristic parameters of wave form B_X wave crest B_F, energy distribution E_F and energy percentage E were extracted in the nodes of Layer 3. The effective evaluation standard was established on the basis of characteristic information extraction. The separability of different defects in time domain eigenvector and frequency domain eigenvector was compared and analyzed respectively. The frequency domain eigenvector with better separability was served as the recognition eigenvalue of classifying defects sorts. BP neural network was used to identify the extracted frequency domain eigenvector, and the total recognition rate reached to 83.3%. Therefore, the method presented in the study is feasible in the wood defects recognition.